Chinese Tech Firms Still Pursue Nvidia AI Chips

0
Nvidia H20 AI chip
  • Despite regulatory pressure, Chinese companies continue to seek Nvidia’s H20 and anticipate the more powerful B30A amid limited local alternatives.

Ongoing Demand Despite Regulatory Pressure

Major Chinese technology firms, including Alibaba and ByteDance, remain interested in acquiring Nvidia’s artificial intelligence chips, according to individuals familiar with procurement discussions. This interest persists despite strong discouragement from Chinese regulators, who have raised concerns about information security risks. Companies are seeking confirmation that their orders for the H20 chip—recently reapproved for sale in China—are being processed. Attention is also turning to Nvidia’s upcoming B30A chip, which is based on the Blackwell architecture and may offer significantly higher performance.

If approved by U.S. authorities, the B30A is expected to cost roughly twice as much as the H20, which currently sells for $10,000 to $12,000. Sources suggest the B30A could deliver up to six times the performance of its predecessor, making the price point attractive to Chinese buyers. Both chips are modified versions of Nvidia’s global offerings, tailored to comply with U.S. export restrictions. These limitations have become a focal point in the broader geopolitical contest over AI and semiconductor dominance.

Strategic Interests and Market Dynamics

China accounted for approximately 13% of Nvidia’s revenue in the previous fiscal year, underscoring the importance of the market despite ongoing tensions. While the U.S. has eased some restrictions, allowing limited exports of downgraded chips, Chinese authorities have summoned firms like Tencent and ByteDance to explain their purchases. No formal ban has been issued, but officials have expressed concern over reliance on American technology. At the same time, the U.S. government has secured a deal to receive 15% of Nvidia’s H20 revenue, adding another layer of complexity to the situation.

Nvidia and other critics of export controls argue that continued use of its chips by Chinese developers helps maintain compatibility with its software ecosystem. This could prevent a full pivot to domestic alternatives such as Huawei and Cambricon. However, Chinese policymakers are actively encouraging the tech sector to reduce dependence on foreign suppliers. The tension reflects a broader struggle between national security priorities and commercial interests.

Performance Gap and Supply Constraints

Despite political pressure, Chinese firms continue to favor Nvidia’s chips due to performance advantages and limited availability of domestic alternatives. Engineers at several companies have noted that Nvidia’s products outperform local offerings in key AI workloads. Supply constraints from Huawei and Cambricon have further reinforced demand for imported solutions. Nvidia’s CEO Jensen Huang has reportedly reassured Chinese partners about the H20’s availability, citing strong demand and ongoing production.

The company holds an inventory of 600,000 to 700,000 H20 units and has asked TSMC to manufacture additional chips. Plans are underway to deliver B30A samples to Chinese clients for testing, potentially as early as September. Nvidia’s recent earnings call revealed that export licenses for the H20 have been granted, though shipments are pending resolution of revenue-sharing terms with the U.S. government. Uncertainty over future sales in China led Nvidia to issue a cautious quarterly forecast, contributing to a 6% drop in its stock value.

Blackwell Architecture and AI Scaling

Nvidia’s Blackwell architecture, which underpins the anticipated B30A chip, represents a significant leap in AI processing capabilities. Designed for large-scale model training and inference, Blackwell chips are optimized for energy efficiency and parallel computation. If the B30A retains core features of this architecture, it could offer Chinese firms a competitive edge in developing advanced AI applications. Given the scale of China’s digital economy, access to such hardware may influence global AI innovation trajectories in the coming years.


Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.